Talks & Speaking
I speak about zero-trust data engineering, non-human identity, clinical de-identification at scale, and what actually breaks when this work runs in production. Venues range from Python conferences to security research workshops to healthcare technology conferences. Publications are listed separately.
Upcoming
Hybrid NLP at Scale: Optimizing Clinical De-identification for High-Throughput FedRAMP Lakes
2nd International Conference on Advancement in Healthcare Technology and Biomedical Engineering (AHTBE 2026)
August 21, 2026 - Vancouver, BC
A hybrid triage architecture for de-identifying unstructured clinical text at petabyte scale: lightweight heuristic scanning clears 90% of low-entropy text at line speed, and only ambiguous, high-context segments get routed to a full Transformer. Benchmarked in a FedRAMP High environment, it matches full-model accuracy (F1 0.98) while cutting compute 40% and ingestion latency 60%.
Build a Zero-Trust Data Pipeline in Python (Without an AWS Account)
PyCon Korea 2026
August 17, 2026 - Seoul, Korea
A three-to-four hour, hands-on tutorial where participants build a small serverless pipeline in plain Python (boto3 and STS) that issues a narrowly-scoped, short-lived credential for every transaction instead of authenticating once and trusting the process for hours. The whole thing runs offline against a local AWS fake (moto), so no AWS account, credentials, or cloud cost, and the second skill people leave with is testing AWS code without AWS. The centerpiece lesson: moto doesn't enforce IAM, so the test guarding your security model passes while the code is broken, and closing that gap is the thing worth remembering. The teaching version of the Identity-Per-Transaction work.
Conference presentations
The Disposable Identity: Eliminating Non-Human Identity Risk in Federal Healthcare Pipelines
USENIX PEPR ‘26
June 1, 2026 - Santa Clara, CA
A production case study of an Identity-Per-Transaction pipeline deployed for a federal life-sciences agency, including the operational realities (latency, concurrency race conditions, debugging credentials that no longer exist) and an honest reckoning with the limits: the identity broker remains a bounded but non-zero root of trust, and eliminating credential-mediated access risk is not the same as solving genomic anonymization.
Zero Trust in 200ms: Implementing Identity-Per-Transaction with Python and Serverless
PyCon US 2026
May 16, 2026 - Long Beach, CA
A production federal life-sciences pipeline that replaces static service accounts with ephemeral Python logic: an identity broker that mints a unique, cryptographically scoped IAM credential for every file transaction and destroys it milliseconds later, a streaming de-identification layer built on Python generators and Microsoft Presidio, and structured logging that produces immutable, audit-ready JSON trails.
Zero-Trust Data Engineering: A Reference Architecture for Serverless, FedRAMP-High Healthcare Pipelines
IEEE BigDataSecurity 2026
May 10, 2026 - New York City, NY
The paper presentation of the Identity-Per-Transaction reference architecture. A serverless zero-trust design (AWS Lambda/Glue) using a just-in-time identity broker and a clean-room streaming de-identification layer, evaluated in production at over 99.9% blast-radius reduction with linear throughput. Proceedings online.
AI as a Collaborative Partner in Data Science Education
AAAI 2025 Spring Symposium - Current and Future Varieties of Human-AI Collaboration
March 31, 2025 - Burlingame, CA
How AI can partner with instructors and students to make data science education more engaging and equitable, from real-time personalized feedback on coding assignments to collaborative student-AI modeling projects, paired with the argument that these gains have to come with deliberate attention to bias, fairness, privacy, and accountability.